Enterprise Knowledge Chatbot. @knowledge-chatbot

The Enterprise Knowledge Chatbot gives your organization a conversational interface over its own approved knowledge — policies, product docs, runbooks, contracts, wikis. Built as a proper RAG system, it retrieves relevant passages, grounds every answer in them, cites its sources, and says so when the knowledge base does not contain an answer. Crucially, it is permission-aware: users only ever get answers derived from documents they are allowed to read. The finish state is the chatbot live with your pilot group, measured for groundedness on a real test set, deployable in as little as 21 days.

Scrums.com builds RAG systems on production-grade vector infrastructure — Pinecone, Weaviate, or pgvector — with ingestion pipelines and re-ranking layers, and tracks hallucination rate as a first-class delivery metric from sprint one. This is that capability, packaged.

Knowledge Ingestion Pipeline

Connectors for your approved sources — SharePoint, Confluence, Drive, Notion, file stores — with chunking, embedding, and scheduled re-indexing so answers track the current documents, not a stale snapshot.

Retrieval & Grounding

Vector retrieval with re-ranking tuned on your corpus, answers constrained to retrieved content with citations, and explicit refusal behavior when retrieval comes back thin.

Permission-Aware Access

Source-system permissions carried through to retrieval — filtered at query time per user — with SSO integration and full audit logging of questions and answers.

Quality Evaluation & Feedback

A test set built from real employee questions, groundedness and accuracy scoring before launch, and in-product feedback loops feeding a tuning backlog.

1. Scope — Fix the knowledge sources, permission model, pilot group, and the evaluation set that defines answer quality.
2. Build — Build ingestion, retrieval, and the chat interface; iterate against the evaluation set until it clears the bar.
3. Handover — Pilot rollout, quality report, and operational docs for keeping sources, permissions, and evaluations current.

The Enterprise Knowledge Chatbot is a menu item on the Scrums.com delivery catalog, available at every plan tier. Add it to your plan backlog and your delivery team schedules it like any other item — scoped, tracked, and reported through the SEOP. See Delivery Plan Tiers.

How do you stop it inventing answers?

Structurally: answers are generated only from retrieved passages, cite those passages, and fall back to an honest refusal when retrieval is weak. Groundedness is then measured on the evaluation set, not asserted.

Does our data end up training a model?

No. Documents are indexed for retrieval, not used for training, and the deployment respects your data-residency and model-hosting constraints — including private or self-hosted models where required.

What does it take to keep it good after launch?

Sources re-index on schedule; the real work is acting on feedback and re-running evaluations as content changes. The LLM Evaluation & Guardrails Package deepens that practice; the Model Ops Retainer keeps it running long-term.